Write *meta* (list of dict) atomically to *path*.

Push dict ➜ actual Blender camera transform.

Get or create a Blender camera object named *name*.

Extract the top 'percentile' bright voxels (or above 'hard_thresh'). We interpret the array shape as (Z, Y, X). index: voxel[z, y, x]

Compute Earth's heliocentric position in ICRS coordinates for a given observation time. Parameters: - obs_time: Astropy Time object for

Scan 'folder' for files named like 'voxel_XXXX.png'. Find the largest XXXX as int, return that + 1. If none found, return 1.

Extract the observation time (DATE-OBS, TIME-OBS) from a FITS file and return as an Astropy Time object. Parameters: - fits_file: Path t

Get the telescope's pointing direction from FITS header RA_TARG and DEC_TARG. Parameters: - header: FITS header containing RA_TARG and D

Cheap test: does *point_xyz* fall inside the camera's view frustum? Horizontal FOV is stored in cam_info["fov"].

Reads a voxel grid from a binary file with the following layout: 1) int32: N (size of the NxNxN grid) 2) float32: voxel_size 3)

Main function: 1. Set up voxel grid and celestial sphere texture. 2. Find and process FITS files (one pass). 3. Analyze voxel grid, f

Duplicate *base_obj_name* around the camera FOV to add clutter.

Process a single FITS image: - Compute Earth's position and telescope pointing. - Read and normalize image data. - Call the C++ funct

Return dict with random position, yaw, pitch, roll, fov (deg).

Try repeatedly until the target is inside the frustum.

Build a rotation matrix (3x3) for Euler angles (rx, ry, rz) in degrees, applied in X->Y->Z order. That is: R = Rz(rz) * Ry(ry) * Rx(rx)